Virtualizing the Network

When we look back a couple of years at how computers evolved we will find that in the beginning of the development of a new resource it was used and addressed directly. A great example is a computers memory. For quite a long time the processor and each program had direct access to every single RAM cell by using each cell's hardware address directly. A program always knew where in memory its code had been stored and there were no differences between cells storing data and cells storing code. I am talking about computers that can be described as 'PCs', others already used new concepts that we can find in today's devices. Over time, direct access became a disadvantage, systems and application programmers did no longer want to take care of where exactly data was stored and that there were no conflicts with other programs or even within a program. So a new layer hiding physical memory from the software was introduced. The microprocessors now had units that translated virtual addresses used by the software to physical addresses. Advantages were that now several programs could use the same addresses to address their data, and the memory management unit translated them to different physical addresses. Also, memory that did not even exist could be simulated, using for instance the hard disc as a media where portions of RAM could be stored temporarily.
Today we can see many similarities when we look at the development of networks. Until a short time ago most computers were not networked. Existing networks were relatively small and clearly laid out, the administrators almost knew each single node personally. We almost always use a host's IP (or its name, which is mapped 1:1 to one or more IPs) to access certain services or programs on that computer, just as we used to access certain regions in memory by their hardware addresses in order to access a certain piece of code (e.g. calling an subroutine) or data. Today we still use memory addresses for this (for instance pointers in C[++]), but they have a completely different meaning. They are virtual addresses rather than physical ones, i.e. we do not know where the data we want to access is physically stored.
In the future a similar process will happen to network addresses: we will no longer address single hosts (cells) directly with their Internet-address (hardware address), but we will use a virtual address (that does not need to be numeric) in order to access a certain service, regardless of what IP (physical address) the host providing the service has. This process has already begun, an example is the Domain Name Service. However, the mapping done by DNS is much less powerful than the one used for virtual memory, so we can assume there is still a lot of potential. In addition, we are still at the beginning of the networking age and we still think in absolute terms when we design new networks. There are concepts, like virtual private networks, virtual LANs or virtual servers, where we can see some of the new directions. Today nobody tries to find or even use the physical address of a memory cell, where some piece of code or data is stored (except for operating systems designers, but that is what they do, providing us with mechanisms that hide those details), but we still almost always use the physical IPs of network nodes, even DNS is hardly more than a mapping name to IP. We could imagine a mechanism just like virtual memory, where we can (virtually) address code, data or services on a network without knowing or even being able to find out what physical address (IP) is associated with the virtual network address. This, however, is beyond this document, all I wanted was to show where NAT may find or has found its place in the entire scheme. NAT will not be the general solution of course, indeed it has been a hack introduced to circumvent limitations of the current system, which is marked by not providing much virtualization of the network yet. NAT is, on the other hand, a useful tool to achieve virtualization on a local scale, as long as more general solutions have not been developed. It is probable that NAT will even contribute to such a general solution (as has become the virtual memory system for virtualizing memory), either by being part of it or by gaining new insights and experiences that will help to understand the virtualization process better, or both.

NAT is being widely used already, although only for purposes described at the beginning, to help with IPv4 address space limitations and it is very successful at that. This is also another reason why I believe IPv6 will not come that fast, the most pressing problem of the IP-addresses has been partly and temporarily solved and many people can live with it. On the other hand recompiling each application for IPv6 is hardly manageable in large networking environments where the administrators have a lot to do anyway.